Archives seem to have come into a kind of fashion in the late 20th century, and increasingly so as of late. There's something knowable, certain, and grounding about the idea of archives, especially in a time of greater existential and perceieved precarity. The training procedures of AI foundation models like GPT have a close relationship with archives too, but instead need their accessibility, volume, and givenness, which allows archives to be composed and instrumentalized as training sets.
Beyond the veil of nostalgia, a home to which one cannot return, this project presents a naive machine learning model alongside an informal archive of music from the early web. These two components have developed alongside each other, and shape each other's sensibilities and expression. The MIDI files collected here were once very new and I've been sifting through them in search of the feeling of transformation, how they unify the possibilties of new aesthetic experiences with the technics of producing and distributing this particular format of media. In both machine learning and the archive, there exist archetypes of transformation and stasis, around which we shore up our hopes and fears about how we ourselves may change.
The neural net model used here is simple by design, and you can check out and run the source code yourself right from your browser. My intent is to have the model be able to express something at once whimsical and general about the underlying archive, and to be simple enough to serve educational purposes. It does not represent the state of the art in 2023, nor is it intended to be used seriously as a tool for music creation. Furthermore, I planned for this model to be portable and cheap to run. In its current implementation, it exists as a ~800k parameter decoder-only transformer model that runs from an AWS Lambda Function every day around noon (GMT) to produce a piece of music around three minutes in length.
This project also owes a debt to BitMIDI, which also developed the MIDI player I've implemented on this site for your listening pleasure.
Before MP3s came to dominate how people would listen to music on the internet, the sounds of the early web (and even BBS and Usenet before the world wide web) were predominantly expressed via MIDI. Its tiny file-size was accomodated by bandwidth limitations of the 1980s and 90s, web-native support for the format came early from browsers like Internet Explorer and Netscape Navigator.
Table of contents for the archive:
This collection of MIDI files is hosted on Dave's J.S. Bach Page, and it covers a portion of John Sankey's MIDI recordings. For a time, he was known as the Harsichordist to the Internet ...
